Show HN: LIA – a self-hosted multi-agent AI assistant (FastAPI and LangGraph) (github.com)

LIA, a self-hosted multi-agent AI assistant built on FastAPI and LangGraph, has made significant strides in AI conversational capabilities. With its latest update, version 1.25.5, LIA has transitioned from a promising prototype to a reliable telephonic assistant, capable of managing calls with features like instant localized greetings, accurate date resolution, and a strict mandate boundary that prevents unrequested commitments. Notably, the assistant incorporates a Human-in-the-Loop (HITL) mechanism with six approval levels to mitigate unpredictable language model costs and control hallucinations, enhancing its reliability for users. The significance of LIA lies in its comprehensive architecture that includes 19+ specialized agents, real-time token tracking, and extensive observability metrics. It offers innovative solutions to common issues faced in AI assistants, such as fragmented integrations and inconsistent performance. Moreover, LIA employs semantic routing for task classification and supports multi-language interactions with full internationalization (i18n), making it versatile for a global user base. The project demonstrates a cutting-edge blend of autonomous execution and human oversight, with an impressive codebase primarily developed by AI, showcasing the potential for advanced AI-driven frameworks in everyday applications.